Face reconstruction method based on supervised learning depth autoencoder

Abstract

A face reconstruction method based on supervised learning depth autoencoder is provided. The method comprises steps of: for a face image with missing or partially shielded information, firstly performing face image feature learning by using the advantages of supervised learning and the efficient feature learning ability of the autoencoder; then, constructing a face reconstruction network with deepstructure by extracting and learning a complete learned feature weight; and finally, reconstructing the complete face information through the constructed face reconstruction network based on deep autoencoder. The method fully combines the advantages of supervised learning and the feature learning ability of the deep autoencoder, extracts the feature of the face image with the missing or partiallyshielded information by using the prior knowledge, reconstructs the extracted features by means of encoding and decoding, repairs missing or shield face information, and has a better result than a traditional method.

Description

technical field [0001] The invention belongs to the field of computer vision, relates to technical methods for face recognition, image restoration, etc., and is specifically a face reconstruction method based on a supervised learning depth autoencoder. Background technique [0002] At present, many face recognition-based applications are based on complete face images, such as security monitoring, automatic access control systems at railway stations, etc. Generally, at this time, the face recognition system can better recognize the face image by using the face reconstruction technology to complete or remove the occlusion of the face image. [0003] Traditional image inpainting methods can be divided into two categories: methods based on partial differential equations and methods based on texture synthesis. Although the two methods are relatively complete in terms of mathematical analysis, there are various assumptions and simplification of the model, so the effect in practic...

AI Technical Summary

Problems solved by technology

[0007] The problem to be solved by the present invention is: in the face recognition task, when the front of the face image is damaged or the face is blocked, the recognition accuracy will be significantly reduced, because the damaged or blocked face in the face recognition process Lack of complete face feature information; most of the existing methods for face reconstruction are to remove noise information in face images, but have no significant effect on face images with missing or occluded frontal information

Images